Diabetes After Hormone Therapy in Breast Cancer Survivors ......breast cancer,4,5 but few have investigated the inverse direction: breast cancer induction of di-abetes. The risk of

JOURNAL OF CLINICAL ONCOLOGY O R I G I N A L R E P O R T

Diabetes After Hormone Therapy in Breast Cancer Survivors:A Case-Cohort StudyRola Hamood, Hatem Hamood, Ilya Merhasin, and Lital Keinan-Boker

A B S T R A C T

PurposeBreast cancer treatments have been associated with an increased risk of multiple health-relatedadverse outcomes, but the relationship with diabetes remains unclear. This study investigated theassociation between hormone therapy and diabetes risk in breast cancer survivors.

Patients and MethodsWe performed a case-cohort study of 2,246 female survivors recruited from the Leumit health carefund who were diagnosed with primary nonmetastatic invasive breast cancer in 2002 through 2012.A 20% random subcohort was sampled at baseline, and all diabetes cases were identified. Adjustedhazard ratios (HRs) with 95% CIs were estimated by weighted Cox proportional hazards regressionmodels.

ResultsOf 2,246 breast cancer survivors, 324 developed diabetes over a mean follow-up of 5.9 years. Thecrude cumulative incidence of diabetes that accounted for death as a competing risk was 20.9%(95% CI, 18.3% to 23.7%). In multivariable-adjusted models, hormone therapy was associated withincreased diabetes risk (HR, 2.40; 95%CI, 1.26 to 4.55; P = .008). The hazard for tamoxifen use (HR,2.25; 95% CI, 1.19 to 4.26; P = .013) was less pronounced than the use of aromatase inhibitors (HR,4.27, 95% CI, 1.42 to 12.84; P = .010).

ConclusionActive hormone therapy is a significant risk factor of diabetes among breast cancer survivors.Although cessation of treatment is not recommended because the survival benefits of hormonetherapy outweigh the risks, preventive strategies aimed at lifestyle modifications may minimize therisk.

J Clin Oncol 36:2061-2069. © 2018 by American Society of Clinical Oncology

INTRODUCTION

Breast cancer has become an increasingly sur-vivable disease probably because of the continuedimprovements in detection interventions andadvancements in treatment.1 However, manysurvive only to be faced with another debilitatingillness that is absent or subclinical at the end oftherapy.2 One such sequel is diabetes. Diabetes isa major contemporary health problem, with es-calating prevalence projected to 552 millionpeople affected globally by 2030.3

Previous studies have suggested diabetes asan independent risk factor for the development ofbreast cancer,4,5 but few have investigated theinverse direction: breast cancer induction of di-abetes. The risk of diabetes may be mediated bycommon shared risk factors such as old age,

obesity,6 insulin resistance,7 sedentary lifestyle,and diet,8 and emerging evidence, althoughlimited by its infancy, points to breast canceradjuvant treatment as potentially leading to thedevelopment of diabetes.9-13 The precise biologicmechanism remains unclear. Weight gain10 andhyperglycemia11 have been proposed in thechemotherapy setting, whereas dysregulations ininsulin secretion associated with estrogen de-pletion and subsequent alterations in glucosehomeostasis9,12 have been suggested as plausiblepathways in the hormone therapy (HT) setting.Estrogens modulate insulin resistance and mayexert a direct protective effect against variousinjuries on the pancreatic b-cell islets that pro-duce insulin. Both insulin resistance and pan-creatic b-cell dysfunction are central disorders inthe pathogenesis of diabetes.14 Hormonal treat-ment with tamoxifen, a selective estrogen receptor

Author affiliations and support information

(if applicable) appear at the end of this

article.

Published at jco.org on April 24, 2018.

Corresponding author: Hatem Hamood,

MD, Leumit Health Services, Haharoshet

13, Karmiel 2165116, Israel; e-mail:

[email protected].

© 2018 by American Society of Clinical

Oncology

0732-183X/18/3620w-2061w/$20.00

ASSOCIATED CONTENT

Appendix

DOI: https://doi.org/10.1200/JCO.

2017.76.3524

DOI: https://doi.org/10.1200/JCO.2017.

76.3524

© 2018 by American Society of Clinical Oncology 2061

VOLUME 36 • NUMBER 20 • JULY 10, 2018

Downloaded from ascopubs.org by Journals Press Access on July 9, 2018 from 162.234.150.177Copyright © 2018 American Society of Clinical Oncology. All rights reserved.

http://jco.org

mailto:[email protected]

http://ascopubs.org/doi/full/10.1200/JCO.2017.76.3524




modulator, or aromatase inhibitors (AIs), which block estrogenproduction, can disrupt the estrogen-insulin interplay and elevatethe risk of diabetes.

Notwithstanding improved survival, breast cancer is con-sidered a leading cause of cancer morbidity and mortality amongwomen worldwide.15 The aggravated risk of diabetes, which cannegatively affect breast cancer prognosis and substantially increaserisk of all-cause mortality,16-18 also worsens this picture. Given thedetrimental impact of diabetes on breast cancer survival, additionalexploration of the role of breast cancer treatment in the devel-opment of diabetes is important not only because it would addvaluable information on the etiology of diabetes but also because itwould help to identify high-risk patients in need of accentuatedclinical care. To date, however, no clear consensus exists aboutwhich breast cancer treatment may induce diabetes. The effect ofchemotherapy, if it exists, tends to be short term,11 whereas theeffect of HT is believed to extend longer.12We hypothesized a priorithat HT is the principal modifier of a possible relation betweenbreast cancer and diabetes and undertook this study to test thishypothesis.

PATIENTS AND METHODS

Study DesignWe performed a case-cohort study nested within a longitudinal

patient-based cohort study that estimated long-term treatment health-related adverse outcomes in breast cancer survivors. Consistent with theapplied case-cohort design, covariate information was collected in all casesand in a representative sample of the parent cohort designated as thesubcohort.19

Source cohort. On the basis of the National Health Insurance Law(1995), all Israeli citizens are insured by one of four health care funds attheir discretion. Study patients were female members of Leumit HealthServices (LHS), a nonprofit Israeli health care fund that covers approxi-mately 10% of the total population. These patients were breast cancersurvivors for at least 1 year and treated for early-stage or regionally ad-vanced invasive breast cancer between January 1, 2002, and December 31,2012. Localized and regional stages were selected to guarantee standardizedtreatment procedures and to avoid differential survival rates that mayinfluence the development of diabetes.9

Patients were considered ineligible if they had in situ or metastaticbreast cancer, had a previous history of any type of cancer, or did notsurvive or were disenrolled from LHS during the first year after diagnosis.A total of 5,142 patients with a diagnosis of female breast cancer (In-ternational Classification of Diseases, Ninth Revision, code 174.x) wereinitially identified from LHS administrative databases. By linkage with theIsrael National Cancer Registry (INCR) and a careful review of LHSelectronic medical records for case ascertainment, 5,085 patients wereconfirmed to have a true diagnosis of breast cancer, of whom 2,644matched the inclusion criteria. A 20% random sample (ie, the subcohort)was selected at baseline.

Diabetes parent cohort. The source cohort was further limited tobreast cancer survivors with no history of diabetes before or during the firstyear after diagnosis. The rationale for the 1-year conditional diabetes-freesurvival was to ensure a sufficient period for commencement of HT. Threehundred ninety-eight women with prevalent diabetes (15%) were ex-cluded. In total, 2,246 breast cancer survivors were included and observeduntil the occurrence of diabetes (index date), death, LHS disenrollment, orthe predetermined censoring date (May 31, 2016), whichever occurredfirst. Time at risk began 1 year after breast cancer diagnosis.

Subcohort and patients. Data processing was restricted to the ran-domly selected subcohort of 448 breast cancer survivors and all who

developed diabetes during the study period (n = 324). A fraction of thediabetes incident cases was part of the subcohort (n = 57). Of the 715eligible candidates for the administration of a constructed questionnaire,145 were not contacted because they had died. The remaining 570 women,representatives of breast cancer survivors who were alive at the time of thestudy (n = 1,780), were contacted, and all completed the survey ques-tionnaire (Fig 1).

Data CollectionLHS pharmacy claims and the diagnoses billing database were used to

obtain information on HT and diabetes, respectively. Information onpotential confounders were elicited from LHS and INCR administrativedatabases along with a questionnaire. Validation of these databases andresolution of discrepancies among sources have been described in detailelsewhere.20

HT exposure. HT was received by 82% of the 570 cases and the sub-cohort within 6 months after breast cancer diagnosis for a median duration of5 years (interquartile range, 4 to 6 years). HTwas started before diagnosis ofdiabetes in all cases, with. 90% receiving treatment at least 1 year before theindex date. However, the treatment was not completed in . 80% of cases bythe time of the diabetes diagnosis. The mean lead time after the index date was1.5 6 0.2 years. The treatment protocol was predominantly based on se-quential therapy (70%) with the use of AIs after 2 to 3 years of tamoxifen tocomplete 5 years of HT; with the exception of five patients who beganAIs (1%)but experienced intolerable adverse effects (as witnessed through review oftheir medical records) and switched to tamoxifen. Only 5% of those treatedwith HT received AIs alone. HT discontinuation was too low (median, 0.08;interquartile range, 0.00 to 0.16 years) to be considered in analyses.

Outcome measure. The primary outcome was diabetes-free survival,which was defined as the time from study entry (1 year after breast cancerdiagnosis) to a first diagnosis of diabetes (International Classification ofDiseases, Ninth Revision, code 250.x) in the LHS electronic encountercodes. The accuracy of the administrative code for diabetes was shown tobe high through a previously validated algorithm that is based on labo-ratory measurements, procurement of diabetes medications, and review ofmedical records.20 The positive and negative predictive values were 94.8%(95% CI, 75.4% to 99.1%) and 100% (95% CI, 96.3% to 100%), re-spectively.20 Of note, diabetes has been used along with five other healthcare areas to assess the quality of community health care in Israel21; thus,health care funds, including LHS, have invested efforts to diagnose thedisease accurately.

Covariates. Information extracted from the INCR included breastcancer diagnosis date, age at diagnosis, immigration status, clinical stage atdiagnosis, type of surgery, and axillary lymph node dissection. Supple-mentary information on region of residence, mean household income(derived from census data on the basis of patient residence), comorbidconditions at time of breast cancer diagnosis, receipt of adjuvant treatment,estrogen receptor status, lymph node involvement, health services utili-zation, and all-cause mortality were abstracted from LHS registries. Thequestionnaire, which was delivered in 2016 only to the subcohort and tobreast cancer survivors diagnosed with diabetes, collected data on eth-nicity, cohabitation status, education, parity, menopausal status at breastcancer diagnosis, family history of diabetes, work history, and lifestyle.

EthicsThe study protocol was reviewed and approved by the institutional

review boards of LHS and the University of Haifa and in accordance withthe Declaration of Helsinki, and all procedures complied with currentIsraeli laws. Patients were fully informed of the study objectives andprocedures, and oral informed consent was obtained from each.

Statistical AnalysisDiabetes excess prevalence relative to the general population was

quantified on the basis of data derived from the National Program for

2062 © 2018 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

Hamood et al


Quality Indicators in Community Healthcare in Israel for the period of2010 to 2013.22-24 We used cumulative incidence function to estimate thecrude incidence of diabetes among the parent diabetes population in thepresence of death as a competing risk.25

To assess the independent relation between breast cancer HT anddiabetes risk, multiple weighted Cox proportional hazards regressionmodels were applied. Impact measures of attributable risk fraction andpopulation attributable fraction (PAF) were assessed.

In a secondary analysis, we tested the effect of HT duration on risk ofdiabetes to address the potential of surveillance bias related to accentuatedcare and increased probability of outcome detection in patients whoinitiated treatment.9 We dichotomized treatment duration using the 1-yearcut point to preserve temporality.

Additional analysis by HTagent was performed. Because agent switchin the sequential setting could occur after the outcome, only the firstreceived agent was considered. Details of the statistical methods used areprovided in the Appendix (online only).

RESULTS

Diabetes prevalence rates in the source population of 2,644 breastcancer survivors increased drastically from 6% in 2002 to 28% in2015 and was above the national norms in 2010 to 2013 (stan-dardized prevalence ratios, 1.61 to 1.81; P, .001). After excludingpatients with prevalent diabetes at baseline (n = 398) and after13 years of observation (mean follow-up, 5.9 years), the crudecumulative incidence rate of diabetes in the presence of deathas a competing risk was 20.9% (95% CI, 18.3% to 23.7%).

A total of 570 subcohort and diabetes cases alive at the time ofthe study were included in the analysis. The mean age at diagnosiswas 55.5 years (SE, 0.5 years), and the mean time elapsed sincediagnosis was 6.4 years (SE, 0.1 years). Overall, participants were

Subcohort noncasesdied and not

contacted(n = 80)

Parent population(n = 2,246)

Prevalent diabetes:Diagnosed before study entry orduring the first year after breast

cancer diagnosis(n = 398)

Cases(n = 324)

Nonsubcohortcases died

and not contacted(n = 53)

Subcohort cases(n = 57)

Subcohort casesdied and not

contacted(n = 12)

Incident breast cancer casesdiagnosed between 2002 and 2012

(n = 2,644)

Subcohort(n = 448)

Subcohort cases(n = 45)

Eligible subcohortcontacted and completed survey

(n = 356)

Eligible cases contacted andcompleted survey

(n = 259)

Fig 1. Flow diagram of study participants.

jco.org © 2018 by American Society of Clinical Oncology 2063

Diabetes in Breast Cancer


http://jco.org

Table 1. Participant Characteristics

Characteristic Diabetes Cases, Weighted % (SE) Noncases, Weighted % (SE)

Weighted HR (95% CI)

Crude Modela P

No. of patients 259 311DemographicMean attained age, years (SE) 63.67 (0.50) 62.76 (0.59) NAEthnicity (Arab v non-Arab)b 9.65 (1.84) 5.47 (1.29) 2.35 (1.17 to 4.72) .017Cohabitation status (unmarried v married)c 37.84 (3.02) 29.90 (2.60) 1.30 (0.89 to 1.90) .172District of residenceNorthern 22.78 (2.61) 20.58 (2.29) 1.50 (0.83 to 2.72) .183Central 29.73 (2.84) 34.41 (2.70) 0.99 (0.57 to 1.73) .984Southern 33.98 (2.95) 28.30 (2.56) 1.20 (0.70 to 2.07) .508Jerusalem 13.51 (2.13) 16.72 (2.12) 1.00 (Reference)

Immigration status (non-Israel born v Israel born) 61.00 (3.03) 52.09 (2.84) 1.11 (0.76 to 1.62) .595Education level, 10 years 57.92 (3.07) 35.05 (2.71) 2.81 (1.69 to 4.66) , .00111-12 years 15.44 (2.25) 15.43 (2.05) 1.63 (0.98 to 2.73) .061$ 13 years 26.64 (2.75) 49.52 (2.84) 1.00 (Reference)

Household income, tertile (median NIS)d

1 (6,284)–lowest 40.54 (3.05) 32.15 (2.65) 1.69 (1.10 to 2.62) .0182 (7,331) 32.82 (2.92) 32.48 (2.66) 1.30 (0.85 to 1.98) .2263 (9,518)–highest 26.64 (2.75) 35.37 (2.71) 1.00 (Reference)

Breast cancer characterizationBreast cancer diagnosis year2002-2005 49.03 (3.11) 27.01 (2.52) 1.42 (0.90 to 2.24) .1342006-2009 35.52 (2.98) 38.59 (2.76) 1.17 (0.74 to 1.85) .5132010-2012 15.44 (2.25) 34.41 (2.70) 1.00 (Reference)

Mean age at breast cancer diagnosis, years (SE) 58.77 (0.49) 54.95 (0.59) NAAge distribution at breast cancer diagnosis, years# 39 0.77 (0.54) 12.86 (1.90) NA40-49 16.60 (2.31) 23.47 (2.41)50-59 41.31 (3.06) 29.58 (2.59)$ 60 41.31 (3.06) 34.08 (2.69)

Regional stage (SEER) 40.54 (3.05) 40.19 (2.78) 1.11 (0.78 to 1.58) .579Lymph node involved 39.77 (3.04) 39.87 (2.78) 1.08 (0.76 to 1.54) .661Estrogen receptor (positive v negative) 83.40 (2.31) 81.60 (2.20) 1.51 (0.98 to 2.32) .064

Breast cancer treatmentSurgeryBreast conserving 80.31 (2.47) 76.53 (2.41) 3.67 (0.37 to 36.13) .266Mastectomy 19.31 (2.45) 21.87 (2.35) 3.25 (0.32 to 32.74) .318No surgery 0.39 (0.39) 1.61 (0.71) 1.00 (Reference)

Axillary lymph node dissection 72.20 (2.79) 66.88 (2.67) 1.22 (0.83 to 1.81) .313Radiotherapy 78.38 (2.56) 78.78 (2.32) 1.10 (0.71 to 1.71) .668Mean radiotherapy duration, days (SE) 36.44 (2.28) 35.09 (0.98) 1.00 (0.99 to 1.02) .546ChemotherapyAnthracycline based 19.31 (2.45) 15.43 (2.05) 0.96 (0.60 to 1.56) .880Taxane based 1.54 (0.77) 3.86 (1.09) 0.58 (0.18 to 1.81) .347Anthracycline and taxane based 30.89 (2.87) 36.33 (2.73) 1.06 (0.70 to 1.61) .782Other 3.47 (1.14) 1.93 (0.78) 2.78 (0.97 to 7.97) .057No chemotherapy 44.79 (3.09) 42.44 (2.81) 1.00 (Reference)

Mean chemotherapy duration, months 3.28 (0.23) 3.82 (0.24) 0.99 (0.96 to 1.03) .679HTAntiestrogens 13.90 (2.15) 25.72 (2.48) 1.37 (0.73 to 2.57) .330AIs 5.02 (1.36) 3.54 (1.05) 1.39 (0.57 to 3.40) .474Antiestrogens and AIs 64.48 (2.98) 52.09 (2.84) 1.57 (1.01 to 2.45) .045No HT 16.60 (2.31) 18.65 (2.21) 1.00 (Reference)

Mean HT duration, years (SE) 4.81 (0.15) 4.10 (0.13) 1.07 (1.01 to 1.14) .031Reproductive and medical historyParity (at least one child birth v nulliparous) 89.58 (1.90) 93.25 (1.42) 0.74 (0.40 to 1.37) .335Mean age at first birth,e years (SE) 23.77 (0.14) 22.90 (0.14) 1.06 (1.00 to 1.12) .061Menopausal status at breast cancer diagnosis

(postmenopausal v premenopausal)79.54 (2.51) 62.70 (2.74) 1.04 (0.59 to 1.84) .889

Mean age at menopause,f years (SE) 49.55 (0.19) 49.13 (0.23) 1.03 (0.96 to 1.09) .439Family history of diabetes 4.25 (1.25) 4.82 (1.22) 0.81 (0.37 to 1.80) .606Comorbidity at time of breast cancer diagnosisHypertension 47.49 (3.11) 29.58 (2.59) 2.38 (1.56 to 3.62) , .001Hyperlipidemia 39.38 (3.04) 32.15 (2.65) 1.18 (0.81 to 1.70) .389Cardiovascular disease 11.97 (2.02) 10.93 (1.77) 1.06 (0.62 to 1.83) .826Depression 3.09 (1.08) 6.11 (1.36) 0.59 (0.26 to 1.37) .221

(continued on following page)


Hamood et al


predominantly non-Arab (93.9%) and married (68.9%). Detailedclinical and demographic characteristics of diabetes cases andnoncases are listed in Table 1. At baseline, diabetes cases were morelikely to be older and less educated than subcohort noncases. Overthe course of follow-up, diabetes cases made more frequentoutpatient visits, consumed more diabetes-promoting drugs (ex-cept for steroids), and received HT for longer periods thannoncases.

Lifestyle components collected after the index date indicatedan unhealthy behavior associated with diabetes characterized byobesity, poor nutrition, and lack of physical activity. Diabetes cases,however, were more successful at controlling their weight thannoncases (Table 2).

In multivariable-adjusted models, HTuse was associated witha significantly greater risk of diabetes than no HTuse (hazard ratio[HR], 2.40; 95% CI, 1.26 to 4.55; P = .008; Table 3). At thepopulation level, the PAF, which is the proportion of diabetes

incidence that could have been prevented among the entire di-abetes cohort had patients not received HT, was 48%. Additionalmodeling of diabetes risk by HT agent revealed significant positiveassociations with tamoxifen and AIs (Table 3). Notwithstandingthat the strength of association and resultant attributable riskfraction were more pronounced with AIs, because of the lowprevalence of AIs among patients with diabetes (6%), which in turnmay explain the relatively wide CIs of the point estimate, theassociated PAF was considerably lower than that of tamoxifen.When HT duration was considered, the risk of diabetes was sig-nificantly elevated during the entire course of HT (Table 3).However, the number of patients in the HT duration interval # 1year or in the AI subgroup was too small for a valid interpretationof the results.

In sensitivity analyses, the significance level of HT in re-lation to diabetes persisted in both the propensity score and themultiple imputation models, whereas the risk estimates were

Table 1. Participant Characteristics (continued)

Characteristic Diabetes Cases, Weighted % (SE) Noncases, Weighted % (SE)


Crude Modela P

EmploymentEmployment before breast cancer diagnosis

Full-time job 45.17 (3.10) 50.48 (2.84) 1.02 (0.64 to 1.63) .920Part-time job 5.79 (1.45) 5.79 (1.33) 0.77 (0.34 to 1.74) .523Retired 9.65 (1.84) 11.58 (1.82) 0.62 (0.36 to 1.09) .095Not working 39.38 (3.04) 32.15 (2.65) 1.00 (Reference)

Employment during breast cancer treatmentFull-time job 1.16 (0.67) 2.57 (0.90) 0.29 (0.07 to 1.18) .084Part-time job 39.00 (3.03) 47.59 (2.83) 0.82 (0.53 to 1.26) .364Retired 9.65 (1.84) 11.58 (1.82) 0.59 (0.34 to 1.02) .061Not working 50.19 (3.11) 38.26 (2.76) 1.00 (Reference)

Workplace reintegration after breast cancer diagnosisg

Same job 83.33 (3.25) 89.14 (2.36) 0.69 (0.32 to 1.47) .339Quit job 16.67 (3.25) 10.86 (2.36) 1.00 (Reference)

Mean absence fromwork because of treatment, weeksh (SE) 15.12 (0.43) 14.27 (0.33) 1.03 (1.00 to 1.05) .045Health service utilizationMean outpatient visits per year, quintile (median)i

1 (11.4)–lowest 15.83 (2.27) 20.90 (2.31) 1.00 (Reference)2 (17.5) 18.53 (2.42) 19.29 (2.24) 0.96 (0.56 to 1.65) .8883 (21.1) 12.74 (2.07) 24.12 (2.43) 1.08 (0.62 to 1.90) .7804 (27.1) 23.55 (2.64) 17.04 (2.13) 2.46 (1.43 to 4.23) .0015 (38.7)–highest 29.34 (2.83) 18.65 (2.21) 1.64 (0.90 to 2.97) .104

Consumption of diabetes-promoting drugsj

Corticosteroids 73.36 (2.75) 81.99 (2.18) 0.57 (0.38 to 0.88) .010Glucocorticoids 57.92 (3.07) 65.59 (2.70) 0.66 (0.46 to 0.94) .022Thiazide diuretics 26.64 (2.75) 12.22 (1.86) 1.89 (1.18 to 3.02) .008b-blockers 44.02 (3.09) 31.83 (2.64) 1.38 (0.95 to 1.99) .091Statins 54.44 (3.10) 38.26 (2.76) 1.32 (0.93 to 1.87) .117Antipsychotics 9.27 (1.80) 9.97 (1.70) 0.81 (0.46 to 1.44) .474

Mean follow-up, years (SE) 3.83 (0.14) 6.81 (0.14) 0.55 (0.50 to 0.61) , .001

NOTE. Weights: cases = 1; noncases = 1/sampling fraction of noncases, where sampling fraction of noncases = subcohort/full cohort without cases = 311/(1,780 2259). Percentages may not sum to 100 because of rounding.Abbreviations: AI, aromatase inhibitor; HR, hazard ratio; HT, hormone therapy; NA, not applicable; NIS, new Israeli shekel.aUnadjusted model; sampling weight: 1,780/356 = 5.00.bNon-Arab women comprised Jews and others.cUnmarried category comprised all women not married or living with a spouse (ie, divorced, separated, widowed, single).dIncomewasmeasured using the proxy variable of mean income of family unit derived from census data on the basis of the patient’s residence and divided into tertiles.eOnly for parous women (cases, n = 232; noncases, n = 290); all parous women had children before diabetes diagnosis.fOnly for postmenopausal women at time of breast cancer diagnosis (cases, n = 206; noncases, n = 195).gOnly for women who worked full-time or part-time jobs before breast cancer diagnosis (cases, n = 132; noncases, n = 175).hOnly for women who worked full-time or part-time jobs during breast cancer treatment (cases, n = 104; noncases, n = 156).iOutpatient visits included all physician visits regardless of specialty measured as average visits per year from date of breast cancer diagnosis until end of follow-up.jDiabetes-promoting drug consumption during follow-up.




http://jco.org

only modestly attenuated (Table 4). Stratification to two periodsto evaluate the possibility of exposure change did not yieldstatistically different results (insignificant test of interaction fortwo independent estimates, z = 20.798; P = .42526), althoughthe 2002 to 2005 estimate was smaller and not significantprobably because it was hampered by the small sample size(Table 4).

DISCUSSION

In this case-cohort study, we demonstrated a positive relationshipbetween HT and diabetes. That HT effects remained significantafter propensity score adjustment, which indicated that selectionbias for HT was not a driving factor for the observed association,

Table 2. Lifestyle Factors Associated With Diabetes

Factor Diabetes Cases, Weighted % (SE) Noncases, Weighted % (SE)


Crude Model* P

No. of patients 259 311Mean BMI, kg/m2 (SE) 30.69 (0.29) 27.45 (0.23) 1.10 (1.06 to 1.14) , .001BMI distribution, kg/m2

Underweight/healthy weight (BMI , 25) 13.90 (2.15) 36.98 (2.74) 0.29 (0.18 to 0.48) , .001Overweight (25.0 # BMI #29.9) 36.29 (2.99) 35.05 (2.71) 0.67 (0.45 to 0.99) .045Obese (BMI $ 30.0) 49.81 (3.11) 27.97 (2.55) 1.00 (Reference)

Weight change since breast cancer diagnosisWeight increased 34.36 (2.95) 46.30 (2.83) 0.97 (0.61 to 1.55) .892Weight decreased 47.88 (3.11) 31.83 (2.64) 1.78 (1.10 to 2.88) .020Same weight 17.76 (2.38) 21.87 (2.35) 1.00 (Reference)

Tobacco useEver 5.41 (1.41) 10.29 (1.72) 0.65 (0.32 to 1.33) .239Former 7.34 (1.62) 9.32 (1.65) 0.87 (0.47 to 1.59) .648Never 87.26 (2.07) 80.39 (2.25) 1.00 (Reference)

Physical activity 44.40 (3.09) 59.49 (2.79) 0.52 (0.36 to 0.74) , .001Physical activity frequency$ 4 times per week 3.09 (1.08) 4.18 (1.14) 0.50 (0.22 to 1.13) .0953 times per week 14.67 (2.20) 23.79 (2.42) 0.44 (0.27 to 0.72) .0011-2 times per week 15.06 (2.22) 23.15 (2.39) 0.50 (0.32 to 0.81) .0041-2 times per month 11.97 (2.02) 9.00 (1.62) 0.68 (0.38 to 1.23) .202# 1 time per month 55.21 (3.09) 39.87 (2.78) 1.00 (Reference)

Healthy diet (No. of fruits and vegetable servings per day)0 19.69 (2.47) 3.54 (1.05) 9.04 (4.32 to 18.93) , .0011-2 69.50 (2.86) 66.56 (2.68) 2.69 (1.71 to 4.22) , .001$ 3 10.81 (1.93) 29.90 (2.60) 1.00 (Reference)

NOTE. Weights: cases = 1; noncases = 1/sampling fraction of noncases, where sampling fraction of noncases = subcohort/full cohort without cases = 311/(1,780 2259). Percentages may not sum to 100 because of rounding.Abbreviations: BMI, body mass index; HR, hazard ratio.*Unadjusted model; sampling weight, 1,780/356 = 5.00.

Table 3. HRs and 95% CIs for Incident Diabetes by Hormone Therapy Exposure and Duration

Hormone Therapy No. of Cases PY* IDR


ARF, % CF PAF, %Crude Model P Adjusted Model† P

Used hormone therapyYes 216 9,029 23.92 1.53 (0.99 to 2.35) .055 2.40 (1.26 to 4.55) .008 58.33 0.83 48.42No 43 2,307 18.64 1.00 (Reference) 1.00 (Reference)

TypeTamoxifen 200 8,534 23.43 1.52 (0.98 to 2.35) .059 2.25 (1.19 to 4.26) .013 55.56 0.77 42.78Aromatase inhibitors 16 495 32.35 1.60 (0.69 to 3.75) .275 4.27 (1.42 to 12.84) .010 76.58 0.06 4.59No use 43 2,307 18.64 1.00 (Reference) 1.00 (Reference)

Duration. 1 year 209 8,658 24.14 1.52 (0.99 to 2.35) .057 2.36 (1.24 to 4.51) .009 57.63 0.81 46.68# 1 year 7 371 18.89 1.66 (0.58 to 4.76) .347 6.48 (1.84 to 22.84) .004 84.57 0.03 2.54No use 43 2,307 18.64 1.00 (Reference) 1.00 (Reference)

Abbreviations: ARF, attributable risk fraction; CF, case fraction (number of exposed cases divided by overall number of cases); HR, hazard ratio; IDR, incidence densityrate (per 1,000 person-years); PAF, population attributable fraction; PY, person-years.*PY lived in the total cohort extrapolated from the random subcohort data; sampling weight, 1,780/356 = 5.00.†Adjusted for ethnicity; residence; cohabitation status; income; education; employment during treatment; chemotherapy type; hypertension; outpatient visits; use ofcorticosteroids, thiazide diuretics, b-blockers, and statins; and year of breast cancer diagnosis. Education, income, outpatient visits, and year of breast cancer diagnosiswere modeled in all analyses as continuous variables to avoid the loss of information. Employment before breast cancer diagnosis, estrogen receptor status, andglucocorticoids were not introduced in all models to avoid multicollinearity.


Hamood et al


together with the finding of persistent diabetes risk throughout theHT period despite intensified outpatient visits among cases buildon and reinforce a previous conclusion by Lipscombe et al,9 whoargued against increased medical surveillance and precipitateddiabetes detection around the time of breast cancer diagnosis andtreatment. Nevertheless, because of the small sample size, thecurrent study results should be replicated in a larger community-based sample.

With respect to hormone agent, the findings substantiateprevious suggestions of tamoxifen as a predominant mediator thatlinks breast cancer to the induction of diabetes,9,12 although themeasure of association in the current sample was to some extentstronger in magnitude. The discrepancy in estimates could beattributed to heterogeneity in HT prevalence, diabetes incidence,or sample size but also could be related to the extent to whichconfounding variables have been considered. We attempted tocontrol for a relatively large constellation of demographic andclinical variables that were occasionally not available in corre-sponding studies, such as menopausal status and parity.

The increased risk associated with AIs observed in this studywas inconsistent with other reports that found no effect9 or evena protective effect12 of these agents. However, one study ac-knowledged that the lack of association between AIs and diabetescould be related to the small proportion of AI users in that study(type II error),9 whereas in the other, no plausible explanation forthe inverse relation was provided,12 which renders the validity ofthese findings questionable, particularly when a breast cancersurvivorship care guideline27 stated, in concordance with thecurrent results, that AIs could raise the risk of diabetes.

Of note, the AI findings observed in this study support theunderlying mechanism related to the effects of estrogen inhibitionthat these studies proposed.9,12 Estrogens play a potential role inthe control of energy balance and glucose homeostasis.14 Estrogendeficiency enhances metabolic dysfunction and predisposes toobesity and progression of the metabolic syndrome, well-established risk factors of diabetes.14,28 Estrogens also may ame-liorate insulin resistance, which is a central disorder in thepathogenesis of diabetes.14 Evidence documented in in vitro andin vivo models29 and subsequently confirmed in humans30 suggeststhat estrogens may regulate insulin secretion by exerting a directprotective effect on pancreatic islets, which augments b-cell hy-perplasia and survival against multiple proapoptotic oxidative and

lipotoxic stimuli.14,31,32 The antidiabetic actions of estrogens havebeen confirmed in randomized controlled trials.33,34 Altogether,estrogens may favorably modify diabetes risk. In contrast to ta-moxifen, which functions as an estrogen modulator throughcompetitive antagonism at its receptor, AIs markedly suppressestrogen levels in postmenopausal women by inactivating thearomatase enzyme responsible for the synthesis of estrogens fromandrogenic substrates35; therefore, their effect on estrogen levelsis expected to be more pronounced, which corroborates ourresults.

The distortion produced by some risk factors of the disease,years of education, and recency of breast cancer diagnosis inparticular led to an underestimation (negative confounding) of thetrue HT-diabetes association, as evident in the weaker crude es-timates compared with the adjusted ones. The majority of studiesin this field, however, have not addressed such risk factors to obtaina more precise estimate of the true association, which in turn maybe a major reason for the inconsistency in results.

The findings of this study should be interpreted within thescope of certain methodological shortcomings. The sample reflectsbreast cancer survivors in one of four health care funds in Israel,which may limit the generalizability of the findings to other set-tings. However, the National Health Insurance Law in Israel allowsall citizens to be registered with and move among any of thesefunds without preliminary conditions, which reduces the potentialfor a selection bias. A potential bias as a result of lack of powerlimits the interpretability of stratified results. However, the cred-ibility of the primary results should be unaffected because theseresults remained robust in a wide range of sensitivity models.Notwithstanding, comprehensive information was collected; re-sidual confounding of uncontrolled factors predisposing to di-abetes, such as lifestyle behavior, could potentially bias the results.Although these factors were available at the time of survey, to avoidpotential reverse causality, they were not accounted for in the riskanalyses. Finally, self-reported data, particularly work history,could be subject to recall bias. However, cancer and treatment arelife-altering events, and women probably would recall their em-ployment status reliably in relation to them.36

With these limits in mind, we believe that the results makepivotal contributions to the limited information currently availableon diabetes and have important implications on breast cancersurvivorship care. Given the high prevalence of diabetes above

Table 4. Sensitivity Analysis for the Association Between Hormone Therapy and Diabetes

Hormone Therapy

Model 1* Model 2†

Model 3‡

2002-2005 2006+

Weighted Adjusted HR(95% CI) P




Yes 1.99 (1.14 to 3.49) .016 1.85 (1.10 to 3.12) .021 1.88 (0.67 to 5.29) .233 3.34 (1.28 to 8.75) .014No 1.00 (Reference) 1.00 (Reference) 1.00 (Reference) 1.00 (Reference)

Abbreviation: HR; hazard ratio.*Model 1: Propensity score model; sampling weight, 1,780/356 = 5.00. Adjusted for propensity score that embeds pretreatment variables (ethnicity, residence,cohabitation status, income, education, chemotherapy type, hypertension, and year of breast cancer diagnosis); employment during treatment; outpatient visits; and useof corticosteroids, thiazide diuretics, b-blockers, and statins.†Model 2: Multiple imputations; sampling weight 2,246/448 = 5.01. Adjusted for all variables listed in Table 3.‡Model 3: Stratification by breast cancer diagnosis year; sampling weights for 2002 to 2005 and 2006+, 554/101 = 5.49 and 1,226/255 = 4.81, respectively. Adjusted forall variables listed in Table 3.




http://jco.org

national standards, the dramatic increase in this prevalence withina relatively brief time, and the establishment of HT as a risk factorthat accounts for . 45% of diabetes incidence at the breast cancersurvivor population level, survivors who receive HTmust be closelymonitored if the estimated reduction in diabetes burden is to beachieved. Practically, HT is amenable to intervention, yet breastcancer survivors should not be denied this treatment because thesurvival benefits of HT outweigh the risks. Focus should beredirected to shared modifiable risk factors, such as lifestyle.Despite current breast cancer survivorship follow-up care plans27

that include recommendations of counseling on healthy lifestylemodification, the issue of regular screening tests for diabetes andimplementation of timely preventive interventions falls betweenthe cracks.

In conclusion, HT is a significant risk factor of diabetes amongbreast cancer survivors. The underlying mechanism is unclear, andadditional research is warranted. Although cessation of treatmentis not recommended and progression of breast cancer often isinevitable, devised strategies aimed at lifestyle modifications in

patients at high risk of diabetes could at least preserve the naturalhistory of breast cancer.

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTSOF INTEREST

Disclosures provided by the authors are available with this article atjco.org.

AUTHOR CONTRIBUTIONS

Conception and design: Rola Hamood, Hatem Hamood, Lital Keinan-BokerProvision of study materials or patients: Ilya MerhasinCollection and assembly of data: All authorsData analysis and interpretation: All authorsManuscript writing: All authorsFinal approval of manuscript: All authorsAccountable for all aspects of the work: All authors

REFERENCES

1. Bodai BI, Tuso P: Breast cancer survivorship:A comprehensive review of long-termmedical issuesand lifestyle recommendations. Perm J 19:48-79,2015

2. Hewitt M, Greenfield S, Stovall E (eds): FromCancer Patient to Cancer Survivor: Lost in Transition.Washington, DC, National Academies, 2006

3. Whiting DR, Guariguata L, Weil C, et al: IDFdiabetes atlas: Global estimates of the prevalence ofdiabetes for 2011 and 2030. Diabetes Res Clin Pract94:311-321, 2011

4. Larsson SC, Mantzoros CS, Wolk A: Diabetesmellitus and risk of breast cancer: A meta-analysis.Int J Cancer 121:856-862, 2007

5. Hardefeldt PJ, Edirimanne S, Eslick GD: Di-abetes increases the risk of breast cancer: A meta-analysis. Endocr Relat Cancer 19:793-803, 2012

6. Wolf I, Sadetzki S, Catane R, et al: Diabetesmellitus and breast cancer. Lancet Oncol 6:103-111,2005

7. Godsland IF: Insulin resistance and hyper-insulinaemia in the development and progression ofcancer. Clin Sci (Lond) 118:315-332, 2009

8. Simon D, Balkau B: Diabetes mellitus, hyper-glycaemia and cancer. Diabetes Metab 36:182-191,2010

9. Lipscombe LL, Fischer HD, Yun L, et al:Association between tamoxifen treatment and di-abetes: A population-based study. Cancer 118:2615-2622, 2012

10. Bordeleau L, Lipscombe L, Lubinski J, et al:Diabetes and breast cancer among women withBRCA1 and BRCA2 mutations. Cancer 117:1812-1818, 2011

11. Lipscombe LL, Chan WW, Yun L, et al: In-cidence of diabetes among postmenopausal breastcancer survivors. Diabetologia 56:476-483, 2013

12. Sun LM, Chen HJ, Liang JA, et al: Associationof tamoxifen use and increased diabetes amongAsian women diagnosed with breast cancer. Br JCancer 111:1836-1842, 2014

13. Juanjuan L, Wen W, Zhongfen L, et al: Clinicalpathological characteristics of breast cancer patientswith secondary diabetes after systemic therapy: A

retrospective multicenter study. Tumour Biol 36:6939-6947, 2015

14. Mauvais-Jarvis F, Clegg DJ, Hevener AL: Therole of estrogens in control of energy balance andglucose homeostasis. Endocr Rev 34:309-338, 2013

15. International Agency for Research on Cancer:Latest world cancer statistics: Global cancer burdenrises to 14.1 million new cases in 2012: Marked in-crease in breast cancers must be addressed, 2013.https://www.iarc.fr/en/media-centre/pr/2013/pdfs/pr223_E.pdf

16. Lipscombe LL, Goodwin PJ, Zinman B, et al:The impact of diabetes on survival following breastcancer. Breast Cancer Res Treat 109:389-395, 2008

17. Barone BB, Yeh HC, Snyder CF, et al: Long-term all-cause mortality in cancer patients with pre-existing diabetes mellitus: A systematic review andmeta-analysis. JAMA 300:2754-2764, 2008

18. Peairs KS, Barone BB, Snyder CF, et al: Di-abetes mellitus and breast cancer outcomes: Asystematic review and meta-analysis. J Clin Oncol29:40-46, 2011

19. Barlow WE, Ichikawa L, Rosner D, et al:Analysis of case-cohort designs. J Clin Epidemiol 52:1165-1172, 1999

20. Hamood R, Hamood H, Merhasin I, et al: Afeasibility study to assess the validity of adminis-trative data sources and self-reported information ofbreast cancer survivors. Isr J Health Policy Res 5:50,2016

21. Jaffe DH, Shmueli A, Ben-Yehuda A, et al:Community healthcare in Israel: Quality indicators2007-2009. Isr J Health Policy Res 1:3, 2012

22. Manor O, Shmueli A, Ben-Yehuda A, et al:National Program for Quality Indicators in CommunityHealthcare in Israel Report, 2008-2010 [in Hebrew].Jerusalem, Israel, School of Public Health andCommunity Medicine, Hebrew University-Hadassah,2012

23. Manor O, Shmueli A, Ben-Yehuda A, et al:National Program for Quality Indicators in CommunityHealthcare in Israel Report, 2010-2012 [in Hebrew].Jerusalem, Israel, School of Public Health and Com-munity Medicine, Hebrew University-Hadassah, 2013

24. Manor O, Shmueli A, Ben-Yehuda A, et al:National Program for Quality Indicators in CommunityHealthcare in Israel Report, 2011-2013 [in Hebrew].

Jerusalem, Israel, School of Public Health andCommunity Medicine, Hebrew University-Hadassah,2014

25. Gooley TA, Leisenring W, Crowley J, et al:Estimation of failure probabilities in the presence ofcompeting risks: New representations of old esti-mators. Stat Med 18:695-706, 1999

26. Altman DG, Bland JM: Interaction revisited:The difference between two estimates. BMJ 326:219, 2003

27. Runowicz CD, Leach CR, Henry NL, et al:American Cancer Society/American Society of Clini-cal Oncology breast cancer survivorship care guide-line. CA Cancer J Clin 66:43-73, 2016

28. Gupte AA, Pownall HJ, Hamilton DJ: Estro-gen: An emerging regulator of insulin action andmitochondrial function. J Diabetes Res 2015:916585,2015

29. Le May C, Chu K, Hu M, et al: Estrogensprotect pancreatic b-cells from apoptosis and preventinsulin-deficient diabetes mellitus in mice. Proc NatlAcad Sci U S A 103:9232-9237, 2006

30. Butler AE, Janson J, Bonner-Weir S, et al:Beta-cell deficit and increased beta-cell apoptosis inhumans with type 2 diabetes. Diabetes 52:102-110,2003

31. Godsland IF: Oestrogens and insulin secre-tion. Diabetologia 48:2213-2220, 2005

32. Liu S, Mauvais-Jarvis F: Minireview: Estro-genic protection of beta-cell failure in metabolicdiseases. Endocrinology 151:859-864, 2010

33. Kanaya AM, Herrington D, Vittinghoff E, et al:Glycemic effects of postmenopausal hormonetherapy: The Heart and Estrogen/Progestin Re-placement Study. A randomized, double-blind,placebo-controlled trial. Ann Intern Med 138:1-9, 2003

34. Margolis KL, Bonds DE, Rodabough RJ, et al:Effect of oestrogen plus progestin on the incidenceof diabetes in postmenopausal women: Results fromthe Women’s Health Initiative Hormone Trial. Dia-betologia 47:1175-1187, 2004

35. Smith IE, Dowsett M: Aromatase inhibitors inbreast cancer. N Engl J Med 348:2431-2442, 2003

36. Short PF, Vasey JJ, Tunceli K: Employmentpathways in a large cohort of adult cancer survivors.Cancer 103:1292-1301, 2005


Hamood et al


http://jco.org

https://www.iarc.fr/en/media-centre/pr/2013/pdfs/pr223_E.pdf

https://www.iarc.fr/en/media-centre/pr/2013/pdfs/pr223_E.pdf

AffiliationsRola Hamood and Lital Keinan-Boker, University of Haifa, Haifa; Hatem Hamood and Ilya Merhasin, Leumit Health Services,

Netanya; and Lital Keinan-Boker, Israel Center for Disease Control, Ramat Gan, Israel.

SupportSupported in part by Grants No. 11658745 and 11658760 from the Council for Higher Education in collaboration with the Graduate

Studies Authority at the University of Haifa.

n n n




http://jco.org

http://www.cancer.net/PREACT

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Diabetes After Hormone Therapy in Breast Cancer Survivors: A Case-Cohort Study

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated. Relationships areself-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For moreinformation about ASCO’s conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/site/ifc.

Rola HamoodNo relationship to disclose

Hatem HamoodNo relationship to disclose

Ilya MerhasinNo relationship to disclose

Lital Keinan-BokerNo relationship to disclose

© 2018 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

Hamood et al


http://www.asco.org/rwc

http://ascopubs.org/jco/site/ifc

Acknowledgment

The results, conclusions, views, and opinions are those of the authors and are not to be construed as the official policy of Israel’sMinistry of Health or of LHS. We thank the advisory committee and all participating breast cancer survivors. We also thank Natan Kahan,PhD, for help and support with data acquisition and study approval.

Appendix

Detailed Statistical MethodsDiabetes excess prevalence relative to the general population was quantified on the basis of data derived from the National

Program for Quality Indicators in Community Healthcare in Israel for the period 2010 to 201322-24 (data on previous or later yearshave not yet been published). Age- and calendar year–standardized prevalence ratios and the corresponding 95% CIs werecalculated assuming a Poisson distribution for the observed diabetes events. We used cumulative incidence function to estimate thecrude incidence of diabetes among the parent diabetes population in the presence of death as a competing risk.25 The oversamplingof cases typical of the case-cohort design was accounted for in the analyses by weighting according to Miettinen (Miettinen O: Am JEpidemiol 103:226-235, 1976), with weights of 1 and inverse subcohort sampling probability given to cases and the subcohort,correspondingly.

Differences in distribution of baseline characteristics between diabetes cases and subcohort noncases were summarized asproportions or means by using the SAS 9.4 (SAS Institute, Cary, NC) survey procedures that incorporate the sample design into theanalyses and were compared using the weighted Cox proportional hazards regression model that included the subcohort and allcases, with age as the time scale and left truncation at study entry. To assess the independent relation between breast cancer hormonetherapy (HT) and diabetes risk, multiple weighted Cox proportional hazards regression models were applied. All clinically andbiologically plausible variables with P# .20 in the univariable analysis and that had no missing values and preceded the index datewere considered for inclusion. Analyses were performed by using an SAS macro adapted from the Monica Risk, Genetics, ArchivingandMonograph project (Kulathinal S, et al: Epidemiol Perspect Innov 4:15, 2007), which computes the weighted estimates togetherwith a robust SE. Deviation from the proportional hazards assumption was detected by both inspecting Schoenfeld-type scaledresiduals of each covariate included in the model and testing the correlation of these residuals with the event time model (Kohl M,et al: Comput Methods Programs Biomed 118:218-233, 2015; Xue X, et al: BMC Med Res Methodol 13:88, 2013). The assumptionof proportionality was violated with cohabitation status. Sensitivity analyses stratified by this variable yielded minimal changes inprimary estimates; therefore, we present results that are based on unstratified models. Impact measures of attributable risk fractionand population attributable fraction using Miettinen’s equation (Miettinen O: Am J Epidemiol 99:325-332, 1974) for adjustedestimates were assessed.

In secondary analysis, we tested the effect of HT duration on risk of diabetes to address the potential of surveillance bias relatedto accentuated care and increased probability of outcome detection in patients who initiated treatment.9 We dichotomizedtreatment duration using the 1-year cut point to preserve temporality. Furthermore, because frequent outpatient visits may bea surrogate for intensive scrutinization for diabetes development, we re-ran the analyses and restricted them to women who hadmore recurrent outpatient visits (fourth and fifth quintiles); the results were similar to the primary analysis, which argues againstdetection bias (data not shown).

Additional analysis by type of HT agent was performed. Because agent switch in the sequential setting could occur after theoutcome, only the first received agent was considered.

Multiple sensitivity analyses were performed to assess the robustness and plausibility of the estimated HT effects, includinga propensity score analysis to assess the possibility of bias that resulted from confounding by indication for HT. The probability ofreceiving HTwas calculated using a logistic regression model that incorporated the patient’s predictive baseline characteristics. Apropensity score was then assigned to each patient according to the probability of receiving HT. The relationship between HTanddiabetes was estimated after substituting the actual covariates in the multivariable model with the composite propensity score asa continuous covariate (D’Agostino RB Jr: Stat Med 17:2265-2281, 1998) and was compared with corresponding results using theindividual covariates with standard modeling.

To address potential introduction of survival bias, because those who died were excluded from analyses, missing values onquestionnaire confounders were imputed by means of the SAS multiple imputation procedure. The monotone pattern was used,and the sample size used for analysis was 715 subcohort and cases (Fig 1). The self-reported variables withmissing informationwereethnicity, cohabitation status, education, and employment during breast cancer treatment. The variables included in the imputation

jco.org © 2018 by American Society of Clinical Oncology



http://jco.org

procedure in the order specified in the variable list were attained age, age at breast cancer diagnosis, year of breast cancer diagnosis,outpatient visits, income, radiotherapy duration, chemotherapy duration, HT duration, residence, immigration status, ethnicity(Israel National Cancer Registry data source), cohabitation status (Leumit Health Services data source), clinical stage, estrogenreceptor status, surgery, axillary lymph node dissection, radiotherapy (ever/never), chemotherapy (ever/never), HT (ever/never),comorbidity at time of breast cancer diagnosis (any of the following: hypertension, hyperlipidemia, cardiovascular disease, de-pression), diabetes, consumption of diabetes-promoting drugs (corticosteroids, glucocorticoids, thiazide diuretics, b-blockers,statins, antipsychotics), ethnicity (questionnaire data source), cohabitation status (questionnaire data source), education, andemployment during breast cancer treatment. Weighted Cox proportional hazards regression models were applied on five imputeddata sets. Results were combined by the MIANALYZE procedure then exponentiated to yield corresponding hazard ratios and 95%CIs and contrasted to a parallel model with missing data.

HT could be subject to change over time in terms of agents applied and duration. To contend with this possibility, analysis wasstratified to two periods, 2002 to 2005 and 2006 to 2012, with 2005 set as the cut point because in this year, an update in thealgorithm for selection of adjuvant therapy for early breast cancer was introduced (Goldhirsch A, et al: Ann Oncol 16:1569-1583,2005). Finally, we repeated the HT agent analyses by restricting the sample to women who did not receive sequential therapy toevaluate the likelihood of exposure misclassification. The results were similar and are not presented. Statistical significance wasdefined as a two-tailed P , .05. All data were analyzed with SAS 9.4 statistical software.

© 2018 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY

Hamood et al


Documents

Diabetes After Hormone Therapy in Breast Cancer Survivors ......breast cancer,4,5 but few have investigated the inverse direction: breast cancer induction of di-abetes. The risk of